Insiders are predicting a giant wave of machine-learning acquisitions as valuations plummet and giants like Snowflake eye takeover opportunities

• As companies' valuations plummet amid the market downturn, insiders and VCs predict an M&A wave.
• Machine-learning startups in the "$10 million ARR club" are prime targets, they say.

Companies' valuations are being cut en masse as the market sours, with many raising down rounds and laying off employees. Insiders and investors are now expecting a big wave of acquisitions, and many machine-learning startups are prime targets.

Some of the startups most likely to get scooped up are part of what investors and insiders sometimes refer to as the "token $10 million ARR club," which refers to companies that picked up a few large initial customers but have yet to break into the mainstream. With a looming downturn, their customers may look to quickly cut costs — which would come at the expense of the startups relying on their deals.

Like with most emerging technology, new machine-learning startups typically find a sweet spot in the rapidly evolving field and build a product as a wedge into customers. Those that find success can then launch additional products and gradually expand until they own a large part of their customers' machine-learning workflows. Those that can't build a sustainable business get acquired or eventually shut down.

With the current market conditions and lower price tags on startups, this means there will likely be plenty of opportunities for acquisitions, either as "acqui-hires" or to buy up key technology. Snowflake, in particular, will likely be eyeing acquisitions after spending $800 million on a machine-learning platform called Streamlit.

"I do think the next six months, if things stay where they are, there could be interesting opportunities on the M&A front. Not necessarily big M&A, but I do think there's going to be some valuation resets on some of the private companies out there that could create interesting opportunities," Snowflake's chief financial officer, Mike Scarpelli, said on the company's most recent earnings call.

Scarpelli went on to explain there were some areas on the company's road map where it may make sense to consider acquisitions for both added staff and tech buys.

"We're not looking for revenue but good teams and technology at a more reasonable valuation," he said.

The same tools spawning billion-dollar valuations lose their luster

The data-catalog space, which includes startups like the $1.2 billion firm Alation and the $5.25 billion company Collibra, is one of several areas in the machine-learning industry that sources say may be challenging to prove as a compelling stand-alone product, which makes it ripe for acquisitions. Another part of that workflow that comes up frequently is feature stores.

Feature stores allow developers to avoid needlessly running massive recalculations when deploying a machine-learning component of a product. The largest player is Tecton, which manages the open-source feature-store tool Feast. Tecton was founded in 2019 by the creators of Uber's Michelangelo machine-learning tools.

Tecton has since moved beyond feature stores to other products, and like many open-source tools, Feast serves as an on-ramp to a more sophisticated (and more lucrative) tool. But insiders question whether a feature store — which at the time was enough to net Tecton $60 million in funding from investors like Sequoia and Andreessen Horowitz — can be a stand-alone product. Both Tecton and Rasgo, another startup that launched on the momentum of feature stores, have since pushed into new areas.

"That terminology has been a little tricky for us. It's really easy to hear the word 'store' and think of a database table," Tecton CEO Michael Del Balso told Insider. "What we've seen is, and we see this again and again, teams who are putting machine learning into production. They underestimate this problem."

It's in many ways a return to the age-old question of whether something is a feature or a product. The machine-learning startup Dataiku, for example, has a feature-store component, while Tecton has rapidly tried to grow beyond feature stores. Both are backed by Snowflake after Tecton raised $100 million earlier this month in a round that also included Databricks.

Snowflake and Databricks could build out the same features these billion-dollar startups have

While Snowflake and Databricks have both bet on Tecton and others, a shadow exists over whether they will launch their own products. Insiders say that as long as they're serving as a way to drive usage of Snowflake and Databricks, they expect the companies to remain supported. But Snowflake and Databricks may also eye certain parts of the workflow, like a feature store, as a component they could add to their own products.

Not all machine-learning startups are in this position. Many investors and insiders have said there are several startups that have likely built enough momentum to avoid being part of a rollup. Hugging Face, recently valued at $2 billion, is one that comes up frequently because of its large community, alongside the experiment-tracking startup Weights & Biases.

At the same time, acquisitions that do crop up may provide lucrative outcomes for investors, who rarely see the lightning-in-a-bottle results of a WhatsApp or Red Hat. Smaller acquisitions and initial public offerings are largely what deliver expected results at a high-enough volume.

"Machine learning is really exciting, but sometimes it's hard to tie to ROI for some of these companies," one insider close to Tecton and other startups said. "Who knows — with this market environment, consolidation might happen really fast."